205 lines
40 KiB
Markdown
205 lines
40 KiB
Markdown
# The terrain-ram coolant-leak audit -- full findings (2026-08-11, ticket #128)
|
||
|
||
> Four-agent audit of the collision->damage->leak chain. VERDICT: pricing,
|
||
> gate, scaling, free floor and sub-hit math are all byte-faithful; the ONE
|
||
> divergence is ELIGIBILITY -- the port read GUID 0x50e590 as the whole
|
||
> HeatSink base family (making myomers/condensers/reservoirs rattle-eligible,
|
||
> every one coolant-bearing -> instant leak per over-floor ram), while the
|
||
> binary's node is the AGGREGATE BANK only: eligible set {HeatSinkBank 0.30,
|
||
> Gyroscope 0.35, Torso 0.25}, 10% of rolls land nowhere, myomers' authored
|
||
> 0.35 weight is DEAD DATA. Gyro/Torso cannot leak; a 1995 ram is gyro/torso
|
||
> damage with at worst a bounded bank drip. Also established: the 0.5-pt free
|
||
> floor eats ALL scrape taps (the per-frame [crashdmg] spam is log noise, not
|
||
> damage -- now sampled), and the +0x268 advancedDamage gate is honored
|
||
> faithfully (night-15's 'leaks' were combat-caused; the collision leak only
|
||
> ever fired where advDmg=1 -- and #162's egg fix would have armed it for
|
||
> every steam match, making this fix urgent). Fixed in mech.cpp
|
||
> (IsDerivedFrom("HeatSinkBank")) same day; bench ram_bench.sh.
|
||
|
||
TICKET #128 — PORT AS-BUILT CHAIN (terrain ram/scrape -> instant coolant leak)
|
||
|
||
=========================================================================
|
||
STAGE 1 — THE EMITTER (physical collision -> TakeDamage dispatch)
|
||
=========================================================================
|
||
The [crashdmg] print lives in Mech::AuthenticGroundAndCollide, game/reconstructed/mech4.cpp:8526. Full path:
|
||
|
||
1a. Per frame, the master mech (only movers with a collisionAssistant run the collision half — replicants bail at mech4.cpp:8427-8428) fetches contacts: GetCurrentCollisions() (mech4.cpp:8431), then Mover::ProcessCollisionList(cols, dt, old_position, &dmg) (mech4.cpp:8450-8452). `Damage dmg` is stack-constructed: Damage::Damage (engine/MUNGA/DAMAGE.cpp:35-45) sets damageAmount=0, burstCount=1; **damageType is NOT initialized by the ctor** — it is stamped Damage::CollisionDamageType(0) inside Mover::ProcessCollisionList at engine/MUNGA/MOVER.cpp:1228, and the accumulated `total_damage` across the frame's contacts is written at MOVER.cpp:1350.
|
||
|
||
1b. Per contact, the engine calls the virtual Mech::ProcessCollision (mech4.cpp:8754-8883, binary @004abb40). It restores frameEntryWorldVelocity first (mech4.cpp:8777 — PORT ADDITION, "COLLISION-DAMAGE ECONOMY GUARD": prevents multi-solid frames compounding StaticBounce reflections; commented as fixing 4x-40x over-pricing), runs the BoxedSolid resolver, then prices via Mover::StaticBounce (mech4.cpp:8796-8797 -> engine/MUNGA/MOVER.cpp:1432-1528).
|
||
|
||
1c. THE AMOUNT FORMULA (engine T0, MOVER.cpp:1527): `return -0.0005 * (vn*vp) * moverMass`, where impact = worldLinearVelocity·normal; damage is 0 if impact > 0 (separating, MOVER.cpp:1465-1469). vn/vp are the reflected normal/parallel components scaled by (1+elasticity). Constants: 0.0005 (MOVER.cpp:1527); minimumBounceSpeed only ZEROES elasticity when -impact <= bounce_min*dt (MOVER.cpp:1470-1473) — **there is NO minimum-speed gate on the damage amount itself; any closing contact prices > 0**. Measured economy (context/combat-damage.md:413-415): raw ≈ 600·v² for a head-on mech (59221 @ |v|=9.94; moverMass is tonnage-scale 60000-90000).
|
||
|
||
1d. OWNER CLASSIFICATION (mech4.cpp:8815-8866): a Mover owner -> mech-vs-mech dispatch via BTDispatchCollisionDamage, edge-gated by ramLastVictim/ramContactLinger 0.35s (mech4.cpp:8837-8846 — PORT ADDITION, explicit one-bump-one-hit; the binary's edge was implicit in its bounce). A CulturalIcon owner -> crunch dispatch + 0.00123f walk-through sentinel. **Plain UnscalableTerrain owners (ground/hills/cliffs/canyon walls) match NEITHER branch — their StaticBounce damage stands with NO edge gate whatsoever** (comment mech4.cpp:8656-8658).
|
||
|
||
1e. THE TERRAIN CRASH RESPONSE (mech4.cpp:8485-8537): if dmg.damageAmount > 0.0f, the frame is fully rejected (velocity zeroed 8487, position restored 8488), [crashdmg] prints (8526-8531, always-on), and an Entity::TakeDamageMessage{inflictor=SELF, zone=-1, dmg VERBATIM} is dispatched at OUR OWN mech (8532-8536, binary @4aa984-4aaab4). Stamps: damageType=0 (from MOVER.cpp:1228), burstCount=1 (Damage ctor), amount=raw StaticBounce kinetic price.
|
||
|
||
1f. FIRE RATE: per frame while scraping. The freshBlock hysteresis was deliberately REMOVED (pricing audit 2026-07-31; comment mech4.cpp:8560-8565). The only natural pacing is the knockdown: iv2 > 40 (i.e. |v| > ~6.3 u/s, const @0x4ab184, mech4.cpp:8540) binds the bmp clip whose cycleSpeed 0 collapses velocity. **Below iv2=40 there is no knockdown, so a sustained scrape with a closing normal component re-prices and re-dispatches EVERY FRAME at full drive speed** (the gait re-derives velocity each frame — context/combat-damage.md:427-431 documents exactly this hazard for the mech-vs-mech case, which got the ramContactLinger fix; terrain never did).
|
||
|
||
=========================================================================
|
||
STAGE 2 — THE DISTRIBUTOR (Mech::DistributeCollisionDamage)
|
||
=========================================================================
|
||
Divert site: mech.cpp:1160-1164 — `if (message->damageData.damageType == 0) { DistributeCollisionDamage(&message->damageData); goto death_tail; }` (binary @0x4a0368; never reaches the zone/armor loop; jumps to the death tail, not return — mech.cpp:1149-1158).
|
||
|
||
Full body mech.cpp:980-1028 (binary FUN_0049ffcc):
|
||
- GATE (mech.cpp:982-985): `player = GetPlayerLink()` (mech+0x190); return if null or !BTPlayerAdvancedDamageOn(player). The bridge (btplayer.cpp:2201-2204) reads **advancedDamageOn2 = the SECOND cell this[0x9a] = +0x268** — it DOES reproduce the binary's +0x268 gate (@0x49ffde) [T1]. Both cells are stamped together at btplayer.cpp:2131 from btMission->advancedDamageOn (mission+0xf0, the egg "advancedDamage" technician flag). NOTE [inference, T3]: commit 3545056 ("steam eggs stamp the host's advancedDamage mission-wide — penalty was silently off for every steam match") means this gate only recently began PASSING in steam matches — plausibly why two testers suddenly report the leak as new.
|
||
- SCALE (mech.cpp:987-990): `scale = ((2000/moverMass) / (100·0.27777779)²) / (1 − elasticity²)` — constants match binary (tbyte @0x4a0148 = 1/3.6, 2000f @0x4a0154; mass +0x20c, elasticity +0x244). ≈ 4.5e-5 at 60t/e=0.2 (measured, context/combat-damage.md:1124). Note the mass CANCELS against StaticBounce's ·mass, so scaled amount ≈ (vn·vp)/(771.6·(1−e²)) — mass-independent.
|
||
- FREE FLOOR (mech.cpp:991-992): scaled < 0.5f -> return (taps free; @0x4a003c). **This floor IS the effective "minimum speed": measured 2.55 pts at 9.6 u/s -> the floor clears at |v| ≈ 4.25 u/s head-on.** Between ~4.3 and ~6.3 u/s a scrape prices 0.5-1.2 pts per frame with no knockdown pacing (see 1f).
|
||
- SUB-HITS (mech.cpp:994-997): n = Round(amount·2), min 1; each sub-hit = amount/n (~0.5 pts each).
|
||
- SELECTION (mech.cpp:1007-1025): per sub-hit, one [0,1) roll walks the full subsystem roster accumulating collisionCriticalHitWeight (+0x10C, authored per subsystem resource "CollisionCriticalHitWeight", mechsub.cpp:200/652); eligibility = IsDerivedFrom HeatSink-family(GUID 0x50e590) / Gyroscope(0x50fdc0) / Torso(0x510b08). Measured weights (scratchpad/SESSION_HANDOFF_20260801.md:39): HeatSinkBank 0.3 / Gyro 0.35 / Torso 0.25 / Myomers 0.35. Un-won rolls land nowhere (weights un-normalized — faithful). Winner takes the sub-hit via MechSubsystem::ApplyDamageAndMeasure (mechsub.cpp:459-481, binary @0x4ac07c = the audited FUN_004ac07c) -> MechSubsystem::TakeDamage (mechsub.cpp:513-538, @0x4ac0bc) -> the subsystem's PRIVATE DamageZone::TakeDamage: `damageLevel += amount × damageScale[0]` (engine DAMAGE.cpp:391), scale = points-normalized from authored damageTypePoints/weaponDamagePoints (mechsub.cpp:220-237).
|
||
- Subsystems per event: up to n (one per sub-hit); a full-run terrain crash (~2.6 pts) = n=5 sub-hits, so up to 5 subsystem rattles per impact.
|
||
|
||
MASTER GATING: **DistributeCollisionDamage is NOT master-gated.** The #150 gate (mech.cpp:1288) covers only the score-report tail and the #162 gate (mech.cpp:1372-1373) only the death tail; the type-0 divert at mech.cpp:1160 runs BEFORE both, on any instance. The port deliberately applies damage on replicants for visual response (comment mech.cpp:1276-1282; binary asserted replicants OUT of this handler entirely, part_012.c:14613-14616). So the shooter-side ram dispatch at the victim's local REPLICANT (BTDispatchCollisionDamage, mech4.cpp:8845; victim->Dispatch, mech4.cpp:8717 — also streamed cross-pod to the master per task #47, messmgr.cpp:348) runs the divert on the replicant too, subject only to the replicant's playerLink/advancedDamageOn2 gate. The TERRAIN self-crash emitter itself is master-only (replicants have no collisionAssistant, mech4.cpp:8427), so #128's chain is master-side.
|
||
|
||
=========================================================================
|
||
STAGE 3 — THE LEAK ONSET (subsystem damage -> coolant leak)
|
||
=========================================================================
|
||
HeatSink::UpdateCoolant, game/reconstructed/heat.cpp:1061-1118 (binary @004adbf8):
|
||
- `coolantDraw = ownZone->damageLevel × heatLoad` (heat.cpp:1078-1080; reads the subsystem's OWN private DamageZone.damageLevel — the exact cell the rattle sub-hit raised).
|
||
- Floors/thresholds (heat.cpp:70-80, byte-verified): draw < 0.0025 -> zeroed (_DAT_0050e3d8); coolantActive flips ON at draw > 0.003 (_DAT_0050e3d4), OFF below 0.0025 — hysteresis (heat.cpp:1110-1117).
|
||
- coolantActive IS the leak: it feeds the ReportLeak attribute (heat.cpp:489, all 19 authored leak watchers -> the "warning: coolant leak" voice) and coolantDraw is published as CoolantMassLeakRate (heat.cpp:481) -> the cockpit LeakGauge (btl4gau2.cpp:984/1008). Coolant then drains at coolantDraw/s with central-tank top-up via DrawCoolant (heat.cpp:1095-1105).
|
||
- heatLoad = 15-sample filtered average of 0.002×temperature×coolantLevel, clamped [0,1] (heat.cpp:70-72, 886-926); typical running value ~0.6 (radiatedHeat ~300 band comment heat.cpp:60-61).
|
||
- THE THRESHOLD: leak springs when damageLevel > 0.003/heatLoad ≈ 0.005 at typical load. A single 0.5-pt sub-hit raises the private zone by 0.5×damageScale[0] ≈ 0.5/points — for any plausible authored points (10-50) that is 0.01-0.05, i.e. **an ORDER OF MAGNITUDE above the leak threshold. One free-floor-clearing sub-hit landing on ANY heat-family subsystem = instant audible leak, deterministically.** And since Myomers is heat-family (weight 0.35) alongside HeatSinkBank/Condensers/Generator/Reservoir, most of the eligible weight IS heat-family; with n=5 sub-hits at run speed, at least one landing heat-family is near-certain — matching "every single time".
|
||
- Design intent note: heat.cpp:1070-1072 states the AUTHENTIC behavior — draw rises only as the heat subsystem takes battle damage; there is no repair, so the leak persists once lit (clears only when heatLoad drops enough to cross the 0.0025 OFF hysteresis — the "warning coo-- war--" oscillation, heat.cpp:876-879).
|
||
|
||
=========================================================================
|
||
GUESSES / STAND-INS / PORT DEVIATIONS in the chain (admitted in comments)
|
||
=========================================================================
|
||
1. mech4.cpp:8868-8873 [T3]: InitialHit(1) vs Slide(2) contact-state split — the two collision floats @0x240/0x244 are fieldAt() STUBS returning 0; port always accumulates InitialHit. (Audio-only; not damage.)
|
||
2. mech4.cpp:8777 + 8436-8441: frameEntryWorldVelocity restore — PORT ADDITION (economy guard), not in the binary (its ground was a heightfield probe, never a collision-list entry).
|
||
3. mech4.cpp:8837-8846: ramContactLinger 0.35s one-bump gate — PORT ADDITION, mech-vs-mech ONLY; the binary's edge was implicit in its bounce separation. **Terrain self-damage got no equivalent gate — the per-frame scrape pricing is the port's own emergent behavior**, bounded in the binary by its velocity-reversing bounce.
|
||
4. mech4.cpp:8566-8567: knockdown mid-clip rebind guard — admitted "cheap belt-and-braces port addition"; the binary has no rebind guard.
|
||
5. mech4.cpp:8508-8510 TODO: throttleState = fall-surface material not forwarded (binary (**[mech+0x2f8])+0x24) — payload nicety, unmapped.
|
||
6. mech.cpp:984: `player == 0` null guard is a port guard (binary reads +0x268 through the link; the audit says the link is always live there).
|
||
7. mech.cpp:1086-1089 [T3]: lastInflictingID write site is a reconstruction guess (binary write site not located).
|
||
8. Replicant damage application (mech.cpp:1276-1282) — deliberate port deviation from the 1995 replicant assertion; consequence: the divert/distributor can run on replicant copies (see master-gating above).
|
||
9. Cross-pod TakeDamage streaming (task #47, messmgr.cpp:348) — port addition; in the 1995 pod, ram damage was network-inert (context/combat-damage.md:416-419).
|
||
10. Subsystem private-zone damageScale values are authored content (res+0x30/+0x44, mechsub.cpp:220-237) — I did not extract the per-subsystem numbers; the "0.01-0.05 per 0.5-pt hit" figure is inference from the points normalization, not a measured value.
|
||
|
||
==============================================================================
|
||
|
||
# Task 4 — Reconstruction archaeology of the collision→coolant-leak chain (#128)
|
||
|
||
## A. TIMELINE — how the chain reached its current shape
|
||
|
||
**2026-07-05 (7b7d465, initial commit).** WinTesla engine baseline. The engine already contained the collision PRICING: `Mover::StaticBounce` returns `-0.0005 * (vn·vp) * moverMass` (C:/git/bt411/engine/MUNGA/MOVER.cpp:1527) and `Mover::ProcessCollisionList` stamps `damage->damageType = Damage::CollisionDamageType` (MOVER.cpp:1228) and accumulates the total into the Damage record. The engine RETURNS the Damage to the caller — it never dispatches it. So the pricing formula is ENGINE truth [T0], not a port invention, and the crash-response FORWARDING is game-side policy.
|
||
|
||
**2026-07-08 (2af401e) — first collision damage dispatch.** `BTDispatchCollisionDamage` added (mech-vs-mech + cultural-icon crunch), unblocked by the STEP-6 cylinder hit-location work. Terrain explicitly did NOT damage: "Terrain (walls/hills) matches neither branch, so it still BLOCKS without damage (faithful to the binary)" — a claim later falsified (the terrain self-damage was hiding in an export gap). The self-TakeDamage on a blocking hit was left as a marked "DEFERRED" note in the response policy (mech4.cpp ProcessCollision header, still visible at mech4.cpp:8660-8664). Era symptom: "grinding has no effect" was structurally true — wall crashes cost nothing.
|
||
|
||
**2026-07-12 (bb795e2) — the ram one-shot + first guard.** First live MP wave found StaticBounce reflections compounding 4-40x across multi-solid frames ("a walking bump one-shot a pristine mech for 112,375 pts"). Fix: `frameEntryWorldVelocity` per-contact restore (port guard for an engine-facility drift, gotcha #16) + `Mech::Reset` motion zeroing. Both are port additions [T3-justified], still in place (mech4.cpp:8436-8441).
|
||
|
||
**2026-07-13 (065c114) — the ×1e-3 normalization [T3, admitted guess].** "Port normalizes x1e-3 to the point economy [T3]" — an INVENTED scale patched in because raw kinetic amounts (~59,000 pts @10 m/s, moverMass ~60-90k) were hitting the victim's ARMOR zones directly (the divert did not exist yet). Also added `ramLastVictim`/`ramContactLinger` contact-edge (one-bump-one-hit, 0.35s re-arm) — a port invention [T3] compensating for the port's kinematic gait velocity never reading as "separating" the way the binary's bounce-reversed physics velocity did. The contact edge SURVIVES today (mech4.cpp:8444-8449).
|
||
|
||
**2026-07-30 (9c83a46, #82 ROOT FIX) — terrain crash SELF-damage recovered.** The wall-grind "knockdown storm / skating" report decomposed; the missing piece was recovered from the EXPORT GAP by raw disasm (scratchpad/night6/gap_4a9770.txt, binary @4aa89f-4aaab4): fallDirection = worldToLocal(damageForce), fallScalar = -(fallDirection·localVelocity), and a self-dispatched `TakeDamageMessage{0x64, zone=-1, engine Damage verbatim}`. This closed the old DEFERRED note — terrain crashes now hurt, per the 1995 binary [T1 raw disasm]. The `[crashdmg]` log line was added here as the pricing arbiter ("run 14 field report: 'tapped a wall and died instantly', so the forwarded amount is suspect vs binary" — mech4.cpp:8523-8531). Also FALSIFIED en route: the action-request-flags drive-suppressor theory. A port-only mid-clip rebind guard kept as "belt-and-braces" (mech4.cpp:8560-8567).
|
||
|
||
**2026-07-30 (14ff351, #83) — the COLLISION DIVERT, hours later.** The self-damage immediately exposed its missing second half: raw kinetic figures were falling into the WEAPON zone/armor loop ("tapped a wall and died instantly"). Recovered @0x4a0368 (hub) → FUN_0049ffcc (distributor, also an export-gap fn, raw disasm scratchpad/night6/gap_49ffcc.txt): gate on victimPlayer advancedDamage copy 2 (+0x268, @0x49ffde); scale = (2000/moverMass)/(100km/h)^2/(1-e²) (tbyte 1/3.6 @0x4a0148, 2000f @0x4a0154); <0.5 FREE (@0x4a015c); Round(2×) sub-hits on random HeatSink-family/Gyro/Torso subsystems via cumulative collisionCriticalHitWeight (+0x10C) and ApplyDamageAndMeasure (@0x4ac07c). All byte-anchored [T1] (mech.cpp:960-1028). Bridges (BTHeatSinkFamily/BTGyroscopeFamily/BTTorsoFamily, BTPlayerAdvancedDamageOn) added same commit.
|
||
|
||
**2026-07-31 (37dd7f9) — collision-pricing audit, two REMOVALS of port inventions.** (A) The 0.4s `gBlockCooldown` contact-hysteresis (a port stand-in) DELETED — it had suppressed the knockdown while contact held, letting the drive re-slam at full price every frame → the night-7 SAURON grind death (110 priced frames @|v|~34.7). Binary needs no guard (crash threshold @0x4ab178 = 0.0 read from the exe). (B) The ×1e-3 normalization DELETED — it double-normalized against the divert's ~4e-5 scale, making victim-side rams a no-op. Raw dispatch restored [T1 @part_012:15324]. Bench: scratchpad/night7/mp_rampricing.sh — 150s wall push = 0 rattle 0 deaths; 9.6 m/s ram = 2.55 pts rattle. KB swept: the "moverMass ≈1.3e6" figure corrected to measured 60-90k tonnage scale (an earlier mis-attribution).
|
||
|
||
**2026-07-31 (a8a0042, #88) — coolant leaks BECOME POSSIBLE.** Until this commit a leak was STRUCTURALLY IMPOSSIBLE: the MechSubsystem ctor filled only its re-declared SHADOW damageZone (#64, gotcha #1), so `HeatSink::UpdateCoolant`'s engine-member read was NULL forever. Fix aliased the engine base member; from this build on, `coolantDraw = zoneDamage × heatLoad` prices live. Same commit documents the authored collision-rattle weights extracted from BTL4.RES [T1]: HeatSinkBank 0.30 / Gyro 0.35 / Torso 0.25 / Myomers 0.35, Condenser+Reservoir 0 for collisions. NOTE the coupling: #83 (rattle) and #88 (leaks) went live ONE DAY apart — the terrain-ram→leak behavior could not have been observed before 2026-07-31 builds.
|
||
|
||
**2026-07-31 (839b040) — night-8: the divert declared "matches the field".** KB records: "RAM damage: the binary's internal-only collision divert MATCHES what players observe (gyro/myomer damage, coolant leak, NO armor damage, grinding inert). May be working as intended; awaiting a player ruling. Gitea #103." (context/open-questions.md:1023-1028). I.e. ram→coolant-leak was seen on night-8 and DISPOSED AS AUTHENTIC-SHAPED, with only the armor question left open.
|
||
|
||
**2026-08-02 (night-10, build 716) — #128 filed.** "Ramming terrain = INSTANT coolant leak (4 for 4 that night). NOTE: still 100%? speed at impact; which terrain." (first tester handout, git show bf04e1c:docs/OPEN_ISSUES_FOR_TESTERS.txt:126-127). #125-#130 appear as "carried" in scratchpad/SESSION_HANDOFF_20260803.md:51. **No commit has ever addressed #128** (git -S/--grep sweeps find only the two handout commits).
|
||
|
||
**2026-08-05 (91bd286) — divert tail corrected.** The #83 block had shipped with an early `return`; the binary actually JUMPS TO THE DEATH TAIL (raw disasm @0x4a0375: jmp 0x4a07b5) — fixed to `goto death_tail` so a rattle death still posts VehicleDead; no score reports on this path (mech.cpp:1149-1164).
|
||
|
||
**2026-08-10 (night-15) — Oracle re-confirms:** "#128 Fail - Happens every single time I collide with terrain at speed, needs to be at some mimimum speed. Confirmed." (scratchpad/night15/OPEN_ISSUES_20260810_commentary2.txt:151). Handout regenerated (40c3594) with "#128 ... Oracle re-confirms STILL LIVE".
|
||
|
||
**2026-08-11 (3545056, #162) — advancedDamage stamping.** "Steam eggs stamp the host's advancedDamage mission-wide (penalty was silently off for every steam match)". Night-14 steam logs show advDmg MIXED across players (docs/KD_CHAIN_AUDIT_20260811.md:194: 8 APPLYING all advDmg=1, 32 SKIPPED all advDmg=0). Since the rattle gate is the SECOND advancedDamage cell (+0x268), set from the same mission egg copy (btplayer.cpp:2128-2132: `advancedDamageOn = advancedDamageOn2 = bt_mission->AdvancedDamageOn()`), the collision rattle was per-node-egg-dependent in steam matches before build 885 and is now uniformly the host's setting.
|
||
|
||
## B. Provenance verdict per element
|
||
|
||
- **Pricing ([crashdmg] amount):** ENGINE (WinTesla MOVER.cpp StaticBounce), shipped 1995 — NOT invented [T0]. The [crashdmg] fields (mass, |v|, force) are diagnostics only.
|
||
- **Terrain crash→TakeDamage forwarding:** 1995 BT game code @4aa89f-4aaab4, recovered by raw disasm 2026-07-30 [T1]. OUR addition history: absent (pre-07-30) → present raw. The engine never forwarded; the game always did.
|
||
- **The divert + DistributeCollisionDamage:** byte-anchored reconstruction [T1], constants read from the exe.
|
||
- **Rattle target weights:** authored data (BTL4.RES) [T1].
|
||
- **Leak pricing (`coolantDraw = zoneDamage × heatLoad`, floor 0.0025 / ON 0.003):** byte-verified constants (_DAT_0050e3d8/_DAT_0050e3d4, heat.cpp:63-80); an earlier 1e-4 misread made them fire 25-30x too eagerly (task #9 correction).
|
||
- **Port inventions still live [T3]:** frameEntryWorldVelocity restore; ramLastVictim/ramContactLinger 0.35s contact edge; the mid-clip bmp rebind guard; the collisionAssistant null guard. Port inventions REMOVED: ×1e-3 normalization, gBlockCooldown.
|
||
- **Known port drift touching the leak:** the heatLoad smoothing window is frame-rate dependent (28 Hz pod vs ~59 fps port halves the window → heatLoad 2x twitchier, heat.cpp:872-880) — documented as the #119 leak-voice stutter cause; also the port's impact |v| comes from the kinematic gait, not binary mover physics (the whole reason the contact-edge guards exist).
|
||
|
||
## C. Prior unresolved observations this audit should explain
|
||
|
||
1. **#103 (night-8, open era question):** ram → gyro/myomer damage + myomer coolant leak, no armor — filed as possibly-authentic, "awaiting a player ruling" on armor. #128 is the terrain-side escalation of the same mechanism; the night-8 disposition ("matches the divert, may be WAI") is the reason nobody treated the leak itself as a bug for ten days.
|
||
2. **The determinism testers report is SHAPED like the reconstruction:** ≥0.5 scaled pts (the "minimum speed" testers observe — ~18.5 u/s at 65t per 37dd7f9's message) → guaranteed ≥1 sub-hit → 0.3+0.35 cumulative weight on heat-family targets → any zone damage × running heatLoad ≥ 0.0025 → leak flag. Whether the 1995 pod also leaked on EVERY over-floor terrain crash is the open question — no ledger records a 1995-side check of leak frequency, only of the divert's structure.
|
||
3. **#135 (leak with NO flashing annunciators) and #97 (three-level leak severity, we only ever produce the lowest)** — both still open, both plausibly entangled with the leak-magnitude side of #128.
|
||
4. **The steam advancedDamage inconsistency:** before 4.11.885 the +0x268 gate varied per node in steam matches; testers could legitimately disagree about whether terrain rams did anything at all. Post-885 it is the host's egg mission-wide — if the host egg authors advancedDamage ON, every player rattles/leaks.
|
||
5. **Un-won roll fidelity:** the distributor deliberately does NOT normalize weights (mech.cpp:976-977 "faithful: the binary does not normalize") — with only heat-family+gyro+torso eligible, a sub-hit can land NOWHERE; nobody has verified the eligible-roster cumulative weight against the binary's actual roster order, which affects how often the rattle hits a heat-family (leaking) member vs fizzling.
|
||
|
||
==============================================================================
|
||
|
||
## Ticket #128 Task 3 — Field receipts from night-14 (build 4.11.857, 4 machines) and night-15 (build 4.11.883, 6 machines) steam logs
|
||
|
||
**Method note:** logs have no per-line timestamps; times were reconstructed from `[rstat] frames=N` anchors at 0.017 s/frame (validated against `[drive] dt=0.017`), linearly interpolated by line number (accuracy ~±1 s). Leak receipt = `[techstat] <sub> condition 2 SET` (condition 2 = CoolantLeaking per game/reconstructed/btl4galm.cpp:116) plus first-seen `[galm-item] cond=2` per subsystem. Scripts: scratchpad crash_analysis{,2,3}.py (session scratchpad dir).
|
||
|
||
### (1) Distribution
|
||
Night 14: **8,528** [crashdmg] events (dave 805, santo 1,065, michael 4,351, rajel 2,307). Night 15: **30,715** (Dave 856, rajel 6,468, acaci 9,378, Michael 9,680, santo 2,765, eleng 1,568). Amounts are wildly bimodal: **~98% of events are amount<2,000** (median ~250–530, p95 ~1.1–1.8k) at scrape speeds |v| 3–6.5; a thin tail of discrete rams: amount 10k–145k at |v| 16–60. n14 amt≥10k: 47 events (0.55%); n15: ~145 (0.47%). |v| shows hard structure: scraping saturates at |v|≈6.4–6.5; a dense cluster at |v|=34.3–34.9 (full-throttle wall rams, esp. rajel: 20+ rams all at 34.x); outliers 40–60 (falls/jump landings — inference). mass field is the mover's own mass (35k–90k, varies per mech pick), never the partner's.
|
||
|
||
### (2) Correlation + empirical threshold — THE KEY RESULT
|
||
The collision→instant-leak signature exists **only in advDamage=1 missions** and **only for discrete high-speed rams**, not scrapes:
|
||
- **Night 14, advDamage=1, amount≥10k: 7 events → 6/7 (86%) followed by an internal-subsystem `[techstat]` receipt within ≤1 s; 4/7 (57%) an explicit CoolantLeaking (cond 2) SET at +0.2–0.4 s.** Michael's 3 rams (14.8k/26.5k/30k raw, |v| 24.4–43.6): ALL THREE → `Myomers condition 2 SET` (coolant leak) at +0.2–0.4 s. Dave's: 76k ram → Gyroscope+HeatSink DMG at +0.0/+0.3 s; 33k ram → Myomers LEAK at +0.3 s. The struck subsystems (Myomers/HeatSink/Gyroscope/Condenser) are exactly the DistributeCollisionDamage lottery families — this is the rattle firing, receipt-latency "instant" [T2 from field receipts].
|
||
- **advDamage=0 (n14 40 events amt≥10k + n15 ~145): ~2/185 (~1%) show a ≤1 s receipt**, and those are ambiguous (one is ERSLaser_1, not a rattle-family sub — concurrent weapon fire). All other follow-ups sit at +1.3–9.9 s, i.e. ambient combat damage.
|
||
- **Empirical threshold:** no event below |v|≈15.5 ever priced ≥8k raw; the smallest ram that sprang a leak was raw=14,811 @ |v|=24.4 (65t). Scrape events (|v|≤6.5) never exceeded ~3.1k raw. So the field threshold ≈ **raw amount ~15k ⇔ |v| in the low-to-mid 20s** — matching Oracle's "needs some minimum speed" and consistent with the 0.5-pt free floor of the (2000/mass)/(27.78)² pricing (raw floor ≈ 0.19×mass = 9.6k–17k depending on tonnage/elasticity) [inference from mech.cpp:987-991 + data].
|
||
- 5-s burst-level correlation over ALL bursts (including scrapes) is only 5–22% (n14) and 0–6% (n15) — the tester "every single time" is about solid rams under advDamage=1, NOT scrapes. Baseline: leaks are common from ordinary weapon fire — 85–100% of cond-2 SETs have NO crashdmg in the prior 5 s (n14: 144/161 orphaned; n15: ~124/142).
|
||
|
||
### (3) Cadence — scraping is priced per frame
|
||
Yes, repeat-fire: consecutive-line [crashdmg] runs of 38–120 events at **25–82 events/s** (frame rate ≈59 → up to ~1.4 pricings per rendered frame). Confirms the santo observation: santo n14 worst run = 50 events in 1.1 s (45/s) totaling 35,601 raw; dave n14 = 50 in 0.6 s (82/s), 51,230 raw; rajel n15 = 94 in 2.2 s, 121,335 raw. Cluster-level (≤2 s gaps): worst = rajel n15, 1,829 events over 43.3 s totaling 1.77M raw. Each per-frame event is individually tiny (med ~300 raw ≈ 0.011 pt after scaling at 70t) → every one falls under the 0.5-pt free floor, so a scrape run costs ZERO through the rattle even when gated on — but the run's raw SUM (25k–121k) would rattle 1–4+ sub-hits if it were priced once as a single impact. The per-frame re-pricing therefore currently under-damages scrapes rather than over-damaging them (through this path).
|
||
|
||
### (4) advDamage=0 cross-check — THE GATE HOLDS
|
||
Night-15 receipts: advDamage=0 for effectively every mission with crash traffic (Dave 11×0/2×1 early, acaci's lone =1 receipt superseded by =0 600 lines later before any ram, Michael's =1 was pre-mission line 56; rajel/santo/eleng all 0). Under those adv=0 receipts, **the instant collision→leak signature is ABSENT** (0–1 borderline in ~145 rattle-eligible rams, 0.7%). The signature appears exactly and only where [exp] advDamage=1 (dave/michael night 14). **Field receipts therefore show the port's rattle path HONORING the +0x268 gate — the leak does NOT come from a gate bypass.** Corollary: the testers who reported "every single time" were almost certainly dave/michael in the n14 advDamage=1 missions (5 and 3 such receipts respectively), or in non-steam sessions outside this evidence set.
|
||
|
||
### (5) Wall vs mech-ram
|
||
Not distinguishable from these receipts: [crashdmg] carries no partner identity and mass= is always the logger's own mover mass. Indirect only: the |v|=34.x pile-up is top ground speed into static geometry; |v| 40–60 outliers imply falls/DFA. No receipt-visible difference in leak outcome by |v| class beyond the amount threshold.
|
||
|
||
==============================================================================
|
||
|
||
TICKET #128 — BINARY TRUTH: THE COMPLETE 1995 COLLISION-DAMAGE CHAIN
|
||
|
||
=== STAGE 1: THE EMITTER (who dispatches, with what amount) ===
|
||
|
||
Pricing is ENGINE code we compile against [T0], dispatch is BT binary code [T1].
|
||
|
||
1a. Per-contact pricing — Mover::StaticBounce (C:/git/bt411/engine/MUNGA/MOVER.cpp:1433-1528) [T0]:
|
||
- impact = worldLinearVelocity · surfaceNormal (the NORMAL component only; grazing scrapes contribute little).
|
||
- impact > 0 (moving away) -> return 0. No other speed gate on damage: minimumBounceSpeed (streamed Mover attr "MinimumBounceSpeed") only zeroes *elasticity for slow hits (MOVER.cpp:1470-1473) — it kills the rebound, NOT the damage.
|
||
- Return value (MOVER.cpp:1527): -0.0005·(vn·vp)·moverMass, which algebraically reduces to raw = 0.0005 · moverMass · (1−e²) · impact² — the kinetic-energy loss in kilojoules (mass in kg, velocity m/s; the DynamicBounce comment at MOVER.cpp:1569 confirms the kJ intent). This is the amount [crashdmg] prints.
|
||
- Mover::ProcessCollision (MOVER.cpp:1365-1414) calls StaticBounce per contact; Mover::ProcessCollisionList (MOVER.cpp:1214-1360) SUMS raw damage across the frame's contacts (total_damage, :1283/:1350), sets damageType = Damage::CollisionDamageType (= 0, :1228) and damageForce = post−pre velocity delta (:1351).
|
||
|
||
1b. Dispatch — the binary's ground+collide response block (port Mech::AuthenticGroundAndCollide, game/reconstructed/mech4.cpp:8383-8537; binary @4aa630-@4aab5f, recovered byte-by-byte from the export gap, raw disasm scratchpad/night6/gap_4a9770.txt) [T1]:
|
||
- crushable-icon contact (0.00123f sentinel) -> gyro crunch only (@4aa7ce-4aa871), NO TakeDamage.
|
||
- blocking contact with damageAmount > 0 -> full frame rejection (velocity zeroed, position restored) + @4aa984-4aaab4 builds TakeDamageMessage{id 0x64, zone=-1, the collision Damage VERBATIM} and Dispatches it at the mech ITSELF. There is NO minimum-amount/speed gate at the emitter: ANY nonzero closing contact dispatches. (iv2 > 40 (u/s)² @0x4ab184 gates only the knockdown/bump ANIMATION ≈ 22.8 km/h.)
|
||
|
||
=== STAGE 2: THE DISTRIBUTOR — FUN_0049ffcc @0049ffcc (part_012.c:14374-14430) [T1, decomp + raw disasm scratchpad/night6/gap_49ffcc.txt] ===
|
||
|
||
Reached ONLY via the hub Mech::TakeDamageMessageHandler @0x4a0230: part_012.c:14636-14638 — `if (damage.damageType == 0) { FUN_0049ffcc(mech, &damage); goto LAB_004a07b5; }` — the goto jumps past the zone/armor loop entirely. Collision damage NEVER touches zone armor, in either gate state.
|
||
|
||
2a. THE GATE (@0x49ffde): `if (*(victimMech+0x190 /*owning BTPlayer*/ +0x268) == 0) return;` — the SECOND advancedDamage cell (egg/technician "splash/collision damage" setting). Gate OFF = TRULY NOTHING: the entire function body is inside this if; combined with the hub's goto, a collision then costs zero armor, zero subsystems, zero anything. [T1]
|
||
|
||
2b. THE SCALING (@0x49ffff-0x4a0030): amount *= ((2000/moverMass) / ((100·(1/3.6))²)) / (1 − elasticity²). Constants from raw disasm: tbyte @0x4a0148 = 0.2777778 (1/3.6, km/h→m/s), 2000f @0x4a0154, 1.0f @0x4a0158, mass @mover+0x20c, elasticity @mover+0x244. Multiplied against the raw kJ from Stage 1, EVERYTHING CANCELS: scaled = 0.0005·m·(1−e²)·v_n² × 2000/(m·771.605·(1−e²)) = (v_n/27.778)² — the squared ratio of normal impact speed to 100 km/h, independent of mech mass and elasticity [T1 algebra on T0+T1 inputs].
|
||
|
||
2c. THE THRESHOLD (@0x4a003c-0x4a0045): scaled < 0.5f (@0x4a015c) -> return, FREE. => minimum damaging impact: v_n ≥ 27.778·√0.5 = 19.64 m/s ≈ 70.7 km/h closing normal speed, for EVERY mech. Walking scrapes (tangential, or slow) are free by construction.
|
||
|
||
2d. SUB-HIT COUNT (@0x4a0051): n = _ftol(scaled × 2.0f (@0x4a0160)) — FUN_004dcd94 is the CRT _ftol, not a helper (Ghidra shows ROUND(ST0); MSVC _ftol truncates — minor rounding uncertainty, n identical at all half-integer boundaries). amount /= n (@0x4a0065). With truncation each sub-hit is 0.5-1.0 points (n=1 for scaled∈[0.5,1), n=2 at 100 km/h, n=3 at ~140 km/h, n=6 at 182 km/h supercharge).
|
||
|
||
2e. TARGET SELECTION (the loop @0x4a006b-0x4a0121; NOT FUN_004ac07c): per sub-hit, roll = FUN_00408050() = rand()·(1/32768) ∈ [0,1) (part_000.c:7930); walk the subsystem roster (count mech+0x124, array mech+0x128) in ROSTER ORDER; an entry is ELIGIBLE iff FUN_0041a1a4(its derivation chain, TAG) matches one of THREE tags (part_012.c:14403-14410): 0x50e590, 0x50fdc0, 0x510b08; accumulate eligible entries' collisionCriticalHitWeight (+0x10C, streamed res+0xDC "CollisionCriticalHitWeight"); first entry whose cumulative ≥ roll receives the sub-hit via FUN_004ac07c and the walk breaks. Weights are UN-NORMALIZED: an un-won roll lands NOWHERE (faithful).
|
||
|
||
2f. ELIGIBILITY — the tags [T1, with a KB-internal contradiction I must flag]: FUN_0041a1a4 (part_002.c:3843) matches by Derivation-object POINTER IDENTITY walked up the parent chain. 0x50fdc0 = Gyroscope's node (DefaultData 0x50fdb0+0x10), 0x510b08 = Torso's node (DefaultData 0x510af8+0x10). 0x50e590 = the HeatSinkBank/AggregateHeatSink (0xBBE) node — NOT the HeatSink base family: decomp-reference.md rows 55+87 (task #57 correction) say so, and it is logically forced by part_013.c:5864 — the HeatSink base ctor itself tests 0x50e590 to let ONLY the bank skip the sink link-attach (heat.cpp:620-628 task #9, live-verified: treating it broader blocked condenser→bank conduction and caused the heat runaway). So the BINARY's eligible set = {HeatSinkBank, Gyroscope, Torso} — authored weights 0.30/0.35/0.25 (Σ=0.90; 10% of rolls land nowhere). The authored Myomer weight 0.35 is DEAD DATA for collisions (myomers are PoweredSubsystem-derived, no gyro/torso/bank tag in their chain; the other +0x10C-like path, zone crits @0049ccc4, uses the ZONE's crit-entry list weights instead). ⚠ CONTRADICTION: game/reconstructed/mech.cpp:1004-1018 (DistributeCollisionDamage) and context/combat-damage.md:1120 read 0x50e590 as the whole HeatSink FAMILY (BTHeatSinkFamily() returns HeatSink::GetClassDerivations(), heat.cpp:1447-1450) — that broadens eligibility to condenser/reservoir/generator/sensor/MYOMERS/weapons, i.e., nearly the entire coolant-bearing branch. This is a prime #128 root-cause candidate (see verdict below).
|
||
|
||
2g. THE APPLICATOR — FUN_004ac07c (part_013.c:4274-4282) [T1]: NOT a picker. It snapshots the target subsystem's PRIVATE zone damageLevel (this[0x38]+0x158, the Subsystem::damageZone @+0xE0), calls the subsystem's TakeDamage virtual (vtbl+0x24) with the (scaled, divided) Damage, returns the delta. The MechSubsystem vtbl+0x24 analog FUN_004ac0bc (part_013.c:4290): zone->TakeDamage (zone vtbl+0x18; DamageZone::TakeDamage @0041e4e0 == DAMAGE.cpp:391: damageLevel += amount × damageScale[type]) then if level ≥ 1.0 (_DAT_004ac140) -> failure states (Destroyed). The private zone's damageScale[0] (collision) is the authored damageTypePoints normalized as scale = 1/(points·defaultArmorPoints) (mechsub.cpp:221-237, ArmourNumerator=1.0 @0x4ac860) — exact authored bank values not pinned [T3].
|
||
|
||
=== STAGE 3: THE LEAK ONSET — HeatSink::UpdateCoolant FUN_004adbf8 (part_013.c:5733) [T1 decomp, constants T2] ===
|
||
|
||
Run per-frame from HeatSink::Update for every HeatSink-DERIVED subsystem only (CLASSMAP.md tree: HeatSink -> Condenser/Reservoir/HeatSinkBank/Generator/PoweredSubsystem -> Sensor/Myomers/MechWeapon...). Gyroscope and Torso are HeatWatcher->PowerWatcher-derived (CLASSMAP.md:34-98) — they have NO coolant members and CANNOT leak.
|
||
- coolantDraw(+0x130) = ownZoneDamage (this[0x38]+0x158) × heatLoad (+0x120, the SMOOTHED radiated-heat running average, attr "NormalizedPressure");
|
||
- draw < 0.0025 (_DAT_0050e3d8) -> zeroed (the floor);
|
||
- the draw drains own coolantLevel(+0x12C) × dt, replenished by DrawCoolant (vtbl+0x38, @0x4add00) recursing up linkedSinks × coolantFlowScale(+0x15C) to the Reservoir (@0x4af3b0) — the cockpit coolant bar visibly drops;
|
||
- HYSTERESIS coolantActive(+0x138): OFF->ON when draw > 0.003 (_DAT_0050e3d4); ON->OFF when draw < 0.0025. coolantActive IS the ReportLeak attribute ridden by the 19 authored leak watchers (the 3-note warning) and the "CoolantLeaking" TechStatusType (@50df17).
|
||
- LEAK CONDITION: zoneDamage × heatLoad > 0.003. One 0.5-1.0-pt rattle sub-hit gives zoneDamage = amount × damageScale[0]; with any working heat load (live port measurements saw myomer heatLoad ≈ 0.014 at idle-ish work: 0.5×0.014 = 0.007 > 0.003) a SINGLE sub-hit plausibly crosses the threshold immediately — the leak is instant once a coolant-bearing subsystem is rattled at all [formula T1; magnitude T3 pending authored bank points].
|
||
|
||
=== THE EXPLICIT ANSWERS ===
|
||
|
||
(a) When does 1995 dispatch damage at all? The self-TakeDamage(type 0) is dispatched on ANY blocking contact with nonzero closing normal speed (no emitter threshold). But it has EFFECT only when the distributor's scaled amount (v_n/27.778 m/s)² ≥ 0.5, i.e. closing NORMAL speed ≥ 19.64 m/s ≈ 70.7 km/h — mass- and elasticity-independent (exact cancellation). Below that every scrape/tap is free; gate (+0x268) OFF makes ALL collisions free. Mech envelope: stat-sheet speeds 40 (gimped)/143 (normal)/182 (supercharged) kph (context/pod-hardware.md:589-598) — so a run-speed head-on ram crosses easily, a walk or graze never does.
|
||
|
||
(b) When does 1995 first spring a coolant leak? Requires ALL of: (i) advancedDamage(+0x268) ON; (ii) v_n ≥ 70.7 km/h (≥1 sub-hit); (iii) the lottery landing on the HeatSinkBank — the ONLY coolant-bearing eligible class — probability 0.30 per sub-hit (gyro 0.35 and torso 0.25 rattles cannot leak); (iv) resulting bank zoneDamage × bank heatLoad > 0.003 (plausibly crossed by one sub-hit on any warm mech). Per-ram leak probability ≈ 1−0.7ⁿ: ~30% at 71 km/h (n=1), ~51% at 100 km/h (n=2), ~66% at 140 km/h (n=3), ~88% at 182 km/h (n=6). 1995 is NEVER "every single time".
|
||
|
||
VERDICT for #128 (inference from the above): the deterministic instant leak the testers report cannot be authentic 1995 behavior at any speed. The port's DistributeCollisionDamage (mech.cpp:1016) broadened 0x50e590 from bank-only to the whole HeatSink family — so myomers (weight 0.35 EACH, several per mech) and other coolant-bearing subsystems enter the lottery, the cumulative weight saturates, and nearly every rattle sub-hit lands on something that runs UpdateCoolant -> leak virtually every qualifying ram. Compounding: the steam eggs now stamp advancedDamage mission-wide (commit 3545056), turning the gate ON for matches where it was silently OFF before, and the run-14 note (mech4.cpp:8523-8525) already flags the forwarded [crashdmg] amount as suspect vs binary.
|